I have A/B tested carousels extensively at my previous job. What I found was that even after a user interacted with the controls, meaning they specifically chose to see subsequent slides, they still clicked the first slide at nearly the same rate as those who never interacted at all (tested with auto slide on and off and variations of the carousels). Intrigued by the finding, I started doing different sorts of the slides (random, most recently added, oldest, etc.) and found no statistical difference. I never had time to figure out why, though, as I took a new position somewhere else. I have a strong suspicion that users see first as most important, though.

Sure, anything could happen. As Shana says:

It is totally possible that there are edge cases where we are wrong. While we’ve personally performed split tests between carousels and their elements and always came up with an individual slide (rather than the whole carousel) as the highest revenue generating source, it is possible that there will be some cases where a carousel is better for conversions.

To go with your example, assuming each group has 50%, it's still tough. The actual conversion rate of the carousel in your example will be T x 60% + (1-T) x 30%, where T is the fraction of users who actually wait for the second image. Some simple algebra suggests that for this to exceed 40%, you'll need T > 1/3.

The problem with your calculation it assumes that people who are likely to get converted by the first image wait around for the second image and get "unconverted".

A good intuitive way to think about it is this: if you're only showing the second carousel image to people who the first image failed to convert then there's zero down-side.

In practice there are some users who would convert if shown the first image for longer who you'll lose by rotating to another image, but that's what you need to measure to decide if it's worth having a carousel.

30% is avg of 20% and 40%. Its the conversion rate of a random person who doesn't stick around long enough to see the second image.

I don't think we really disagree - you should test the carousel, and cases where it wins will be edge cases usually.

the downside is they may bounce instead of waring out the time.

You can get the same set of bites faster and more effectively by simply placing both images directly onto the page. This is why Amazon Kindle and Apple product pages tend to be so tall. People can scroll a lot faster than they can wait for a carousel to go through.

This.

I've personally never seen carousels out-perform alternatives like having multiple images on the page, or multiple landing pages with separate messaging, or just dropping multiple options for the one that's most important.

I'm sure that there are some cases where they're an effective option - but I've yet to come across one in testing.

As Brad Frost once observed "Carousels exist to keep people from beating the shit out of each other in meetings." - https://twitter.com/karenmcgrane/status/316563205294010370

And sometimes "people" is just the different people in your head who want to talk about ALL THE THINGS - when talking about just the most important one is what you really need to do.

This assumes one important problem that sees to not be common - that the person viewing the carousel will stick around long enough to view the next choice (or all the choices)

In an ideal world, what you should do instead is parametric optimization of some sort instead of a carousel - so you can hit that other 20%